search

CUMLSec / XDA

78 stars 13 forks source link

Is the finetune dataset complete or just samples? #9

Open Escapingbug opened 2 years ago

Escapingbug commented 2 years ago

Thanks for releasing the project. I'm currently working on some experiment with this project but found that unable to reproduce the result with the finetune data currently provided.

I prepared my own data with the same raw data (well, gcc part) but my experiment result was awful. Think that if I can reproduce the finetune part, I might some insights of how to adjust my approach.

Hope I can get your help!

peikexin9 commented 2 years ago

Thanks for your interest, @Escapingbug. Are you preparing your own finetuning (train/test) set? I have helped several people finetuning on their provided dataset and always see very good results (with different compilers and other settings). Maybe you could share with me what you have and how you set up the training?

Escapingbug commented 2 years ago

Thanks for you kindness. I will clean up my code and do some test. If it still works I will post that here.